| |

Strict Mode Fix

The capability matrix was showing the wrong value for Bounce Tracking Protection in Strict mode. The bug was in how feature codes get resolved.

Bug 2020404 | D286581

The Problem

The ETP feature string in firefox.js encodes which features are enabled in Strict mode as comma-separated codes like tp,tpPrivate,cookieBehavior5,cm,fp,stp,btp,lna. For integer prefs, the code usually includes the value as a suffix: cookieBehavior5 means โ€œset network.cookie.cookieBehavior to 5.โ€

Bounce Tracking Protection (BTP) broke this pattern. Its feature code is just btp with no numeric suffix, but its pref (privacy.bounceTrackingProtection.mode) is an integer. The standard default is mode 3 (dry-run/observing), and Strict should set it to mode 1 (enabled). But _resolve_strict_value was falling through to the generic integer-suffix parsing path, finding no suffix, and returning the standard default (3) instead of the correct Strict value (1).

The Fix

Added a named case for the btp feature code:

def _resolve_strict_value(feature, strict_features, standard_value):
    code = feature["code"]
    enabled = code in strict_features
    pref_value = str(standard_value)

    if pref_value in ("true", "false"):
        return "true" if enabled else "false"

    # BTP uses a named mode integer, not a numeric suffix in the feature code.
    # MODE_ENABLED=1 is the active state; strict enables it by setting mode to 1.
    if code == "btp":
        return "1" if enabled else standard_value

    # For integer prefs (cookieBehavior), the feature code name encodes the
    # enabled value. E.g. "cookieBehavior5" means set to 5.
    ...

I also updated the description for Local Network Access (LNA) to note that itโ€™s gated behind an additional Nimbus pref (network.lna.etp.enabled), which the feature string alone doesnโ€™t capture.

What I Learned

  1. Convention-based parsing breaks on exceptions. The numeric-suffix convention for feature codes worked for most features, but BTP was the exception. When you build a parser around a convention, you need explicit handling for the cases that donโ€™t follow it.
  2. Feature flags can have multiple layers of gating. LNA is โ€œenabledโ€ in the ETP Strict string, but also requires a separate Nimbus experiment flag. The capability matrix needs to reflect both layers to be accurate.